Predicting the structure and function of a public system without user input is fraught with danger, but this user input is very difficult to obtain without a working prototype. Traditional system design methods make little provision for evaluation and testing by public users. Weaving the results of evaluation and testing into the end of the system development process presents risks: how, for example, can we address unforeseen navigation problems identified during user evaluation, while at the same time ensuring that tight deadlines are met? Another challenge is that we often work with teams of specialists (interface designers, HTML developers, database specialists, content creators etc): alteration of one element often has knock-on effects on other team members. Drawing on experience with the COMPASS project at the British Museum, the author will present practical ideas for incorporating evaluation into the system development process.